AI agent mesh is the infrastructure pattern enabling fleets of autonomous AI agents to discover, communicate, and collaborate at enterprise scale — forming a governed digital workforce rather than isolated bots. Organizations adopt it because isolated agents cannot solve cross-functional problems or safely handle mission-critical workflows without centralized governance, identity, and observability. The critical insight is that an agent mesh is fundamentally a distributed-systems problem first and an AI problem second: every pattern from service meshes — identity, routing, circuit breakers, tracing — applies, but is complicated by non-deterministic, long-running, and autonomous behavior.
What This Cheat Sheet Covers
This topic spans 16 focused tables and 128 indexed concepts, 124 flashcards. Below is a complete table-by-table outline of this topic, spanning foundational concepts through advanced details.
A jump-to index of every table row in this cheat sheet.
An interactive map of every table and concept in this topic.
Table 1: Agent Mesh — Core Concepts and Architecture
The five-layer Agent Mesh reference model separates intelligence (the Agent Fabric) from governance (the AI Control Plane) so each can evolve independently. Understanding these layers is the prerequisite for architecting a system that scales without becoming ungovernable.
| Concept | Example | Description | |
|---|---|---|---|
Multiple specialized agents (billing, support, compliance) collaborating on a customer-escalation workflow | Infrastructure pattern connecting multiple AI agents so they can discover, communicate, and coordinate across organizational and vendor boundaries. | ||
MuleSoft Flex Gateway enforcing PII guardrails on every agent action | • The mandate layer (Layer 5 of the mesh stack): enforces security, compliance, and operational guardrails • hosts orchestration, circuit breakers, and inference auditing | ||
Agent Registry + LLM Gateway + adaptive routing serving 50 specialized agents | • The intelligence layer (Layer 4): centralizes agent management, routing, and LLM-agnostic execution • prevents agent sprawl via a registry | ||
API gateway, event bus, and RAG engine bridging agents to ERP and CRM systems | • Connects agents to systems of record (Layer 3) • agents consume existing integrations rather than replace them | ||
Dozens of duplicate, untracked agents running with production credentials after a hackathon | • Uncontrolled proliferation of agents without centralized registry, audit logs, or access policies • primary governance failure in enterprise AI | ||
plan = llm.think(goal)result = tools.run(plan)next = llm.reflect(result) | The iterative operational loop of autonomous agents: plan (decompose goal), execute (invoke tools), reflect (assess progress and replan). | ||
JSON at /.well-known/agent.json listing skills, endpoint, and auth schemes | • Machine-readable agent manifest enabling zero-config discovery • comparable to an OpenAPI spec for an agent rather than an API | ||
Governance board owning agent policies, risk thresholds, and escalation paths | • Organizational framework for governing autonomous AI at scale • defines ownership, accountability, and control-layer constraints on agent autonomy |